I don't get this because we've seen decision making under pressure, for judges. The conclusions are invariably that even the most reliable indicators are violated occasionally (human intelligence is of course a multi-armed-bandit algorithm so this is exactly what you'd expect, and want). Whether the people making decisions agreed they were experimenting or not did not matter at all. They also weren't able to explain this (and even when they did, they just provided the explanation one of the lawyers provided, which sounds fine, until you realize that of course
two lawyers provided an explanation. So they are giving exactly the wrong information to understand the decision: judge chooses between "A, because ..." and "B, because ...", and at that point your decision is about the difference between A and B, not about A or B themselves)
Whilst I have not seen a comparable study with doctors' decision making ... I have very little doubt it's exactly the same. Doctor's AREN'T reliable and are in fact, no matter how hard they deny it, are experimenting on patients and learning from it.
Of course, when you think about more than a single patient, this is what you want. With this I mean you want (ideally many small, some large) mistakes to happen as long as it isn't you under the knife, but you're expecting at some future time it'll be you, because that way the doctors will maximally improve by then. Once you're under the knife, I guess you don't want to see experiments, but yeah, there's a price for skill.
And there's a price to be paid in mistakes for human skill as well. Even for doctors. Even for surgeons.